BEGIN {
$MINNAMELEN = 25;
$PROGRAMNAME = 'codeml' . ($^O =~ /mswin/i ?'.exe':'');
if( defined $ENV{'PAMLDIR'} ) {
$PROGRAM = Bio::Root::IO->catfile($ENV{'PAMLDIR'},$PROGRAMNAME). ($^O =~ /mswin/i ?'.exe':'');;
}
%VALIDVALUES = (
'outfile' => 'mlc',
'noisy' => [ 0..3,9],
'verbose' => [ 1,0,2],
'runmode' => [ -2, 0..5],
'seqtype' => [ 1..3],
'CodonFreq' => [ 2, 0,1,3,4,5,6,7],
'aaDist' => [ 0,'+','-', 1..6],
'aaRatefile' => 'wag.dat',
'model' => [0..3,7],
'NSsites' => [0..13],
'icode' => [ 0..10],
'Mgene' => [0,1],
'fix_kappa'=> [0,1], 'kappa' => '2', 'fix_omega'=> [0,1], 'omega' => '1', 'fix_alpha'=> [1,0], 'alpha' => '0.', 'Malpha' => [0,1], 'ncatG' => [1..10],
'clock' => [0..3],
'getSE' => [0,1],
'RateAncestor' => [1,0,2],
'Small_Diff' => '.5e-6',
'cleandata' => [0,1],
'ndata' => 1,
'method' => [0,1],
'fix_blength' => [0,-1,1,2],
);} |
The rest of the documentation details each of the object methods.
Internal methods are usually preceded with a _
Valid and default values for codeml programs are listed below. The
default values are always the first one listed. These descriptions
are essentially lifted from the example codeml.ctl file and pamlDOC
documentation provided by the author.
CodonFreq specifies the equilibrium codon frequencies in codon
substitution model. These frequencies can be assumed to be equal (1/61
each for the standard genetic code,
CodonFreq = 0), calculated from
the average nucleotide frequencies (
CodonFreq = 1), from the average
nucleotide frequencies at the three codon positions (
CodonFreq = 2),
or used as free parameters (
CodonFreq = 3). The number of parameters
involved in those models of codon frequencies is 0, 3, 9, and 60
(under the universal code), for
CodonFreq = 0, 1, 2, and 3
respectively.
aaDist specifies whether equal amino acid distances are assumed (=
0) or Grantham's matrix is used (= 1) (Yang et al. 1998).
runmode = -2 performs ML estimation of dS and dN in pairwise
comparisons. The program will collect estimates of dS and dN into the
files 2ML.dS and 2ML.dN. Since many users seem interested in looking
at dN /dS ratios among lineages, examination of the tree shapes
indicated by branch lengths calculated from the two rates may be
interesting although the analysis is ad hoc. If your species names
have no more than 10 characters, you can use the output distance
matrices as input to Phylip programs such as neighbor without
change. Otherwise you need to edit the files to cut the names short.
model concerns assumptions about the dN/dS rate ratios among
branches (Yang 1998; Yang and Nielsen 1998).
model =0 means a single
dN/dS ratio for all lineages (branches), 1 means one ratio for each
branch (free ratio model), and 2 means arbitrary number of rations
(such as the 2-ratios or 3-ratios models. with
model =2, you may
specify the omega ratios for the branches using branch labels (read
about the tree structure file in the document). This option seems
rather easy to use. Otherwise, the program will ask the user to input
a branch mark for the dN/dS ratio assumed for each branch. This should
be an integral number between 0 to k - 1 if k different dN/dS ratios
(omega_0 - omega_k - 1) are assumed for the branches of the
tree.
Bioperl note basically, doing this interactively is not going
to work very well, so this module is really focused around using the 0
or 1 parameters. Read the program documentation if you'd like some more
detailed instructions.
NSsites specifies models that allow the dN/dS ratio (omega) to vary
among sites (Nielsen and Yang 1998, Yang et al. 2000)
Nssites = m
corresponds to model Mm in Yang et al (2000). The variable
ncatGis used to specify the number of categories in the omega distribution
under some models. The values of ncatG() used to perform our
analyses are 3 for M3 (discrete), 5 for M4 (freq), 10 for the
continuous distributions (M5: gamma, M6: 2gamma, M7: beta, M8:beta&w,
M9:beta&gamma, M10: beta&gamma+1, M11:beta&normal>1, and
M12:0&2normal>1, M13:3normal>0). This means M8 will have 11 site
classes (10 from the beta distribution plus 1 additional class). The
posterior probabilities for site classes as well as the expected omega
values for sites are listed in the file rst, which may be useful to
pinpoint sites under positive selection, if they exist.
To make it easy to run several
Nssites models in one go, the
executable
Bio::Tools::Run::Phylo::PAML::Codemlsites can be used,
which asks you how many and which models to run at the start of the
program. The number of categories used will then match those used in
Yang et al(2000).
As noted in that paper, some of the models are hard to use, in
particular, M12 and M13. Recommended models are 0 (one-ratio), 1
(neutral), 2 (selection), 3 (discrete), 7 (beta), and 8
(beta&omega ). Some of the models like M2 and M8 are noted to be
prone to the problem of multiple local optima. You are advised to run
the program at least twice, once with a starting omega value <1 and a
second time with a value >1, and use the results corresponding to the
highest likelihood. The continuous neutral and selection models of
Nielsen and Yang (1998) are not implemented in the program.
icode for genetic code and these correspond to 1-11 in the genbank
transl table.
0:universal code
1:mamalian mt
2:yeast mt
3:mold mt,
4:invertebrate mt
5:ciliate nuclear
6:echinoderm mt
7:euplotid mt
8:alternative yeast nu.
9:ascidian mt
10:blepharisma nu
RateAncestor For codon sequences, ancestral reconstruction is not
implemented for the models of variable dN/dS ratios among sites. The
output under codon-based models usually shows the encoded amino acid
for each codon. The output under "Prob of best character at each node,
listed by site" has two posterior probabilities for each node at each
codon (amino acid) site. The first is for the best codon. The second,
in parentheses, is for the most likely amino acid under the codon
substitution model. This is a sum of posterior probabilities across
synonymous codons. In theory it is possible although rare for the most
likely amino acid not to match the most likely codon.
Output for codon sequences (seqtype = 1): The codon frequencies in
each sequence are counted and listed in a genetic code table, together
with their sums across species. Each table contains six or fewer
species. For data of multiple genes (option G in the sequence file),
codon frequencies in each gene (summed over species) are also
listed. The nucleotide distributions at the three codon positions are
also listed. The method of Nei and Gojobori (1986) is used to
calculate the number of synonymous substitutions per synonymous site
(dS ) and the number of nonsynonymous substitutions per nonsynonymous
site (dN ) and their ratio (dN /dS ). These are used to construct
initial estimates of branch lengths for the likelihood analysis but
are not MLEs themselves. Note that the estimates of these quantities
for the a- and b-globin genes shown in Table 2 of Goldman and Yang
(1994), calculated using the MEGA package (Kumar et al., 1993), are
not accurate.
Results of ancestral reconstructions (
RateAncestor = 1) are collected
in the file rst. Under models of variable dN/dS ratios among sites (NSsites models),
the posterior probabilities for site classes as well as positively
selected sites are listed in rst.
INCOMPLETE DOCUMENTATION OF ALL METHODS